Examples with IndexedPixel qupath.opencv.tools.OpenCVTools.IndexedPixel used on opensource projects

Example 1 with IndexedPixel

use of qupath.opencv.tools.OpenCVTools.IndexedPixel in project qupath by qupath.

the class DensityMaps method findHotspots.

/**
 * Find hotspots in a density map.
 *
 * @param hierarchy hierarchy used to obtain selected objects and add hotspots
 * @param densityServer the density map to query
 * @param channel channel in which to find hotspots (usually 0)
 * @param nHotspots maximum number of hotspots to find per selected annotation
 * @param radius hotspot radius, in calibrated units
 * @param minCount minimum value required in the 'count' channel (the last channel)
 * @param hotspotClass the classification to apply to hotspots
 * @param deleteExisting optionally delete existing annotations identified as hotspots
 * @param peaksOnly optionally restrict hotspots to only include intensity peaks
 * @throws IOException
 */
public static void findHotspots(PathObjectHierarchy hierarchy, ImageServer<BufferedImage> densityServer, int channel, int nHotspots, double radius, double minCount, PathClass hotspotClass, boolean deleteExisting, boolean peaksOnly) throws IOException {
    if (nHotspots <= 0) {
        logger.warn("Number of hotspots requested is {}!", nHotspots);
        return;
    }
    logger.debug("Finding {} hotspots in {} for channel {}, radius {}", nHotspots, densityServer, channel, radius);
    Collection<PathObject> parents = new ArrayList<>(hierarchy.getSelectionModel().getSelectedObjects());
    if (parents.isEmpty())
        parents = Collections.singleton(hierarchy.getRootObject());
    double downsample = densityServer.getDownsampleForResolution(0);
    var toDelete = new HashSet<PathObject>();
    // Handle deleting existing hotspots
    if (deleteExisting) {
        toDelete.addAll(hierarchy.getAnnotationObjects().stream().filter(p -> p.getPathClass() == hotspotClass && p.isAnnotation() && p.getName() != null && p.getName().startsWith("Hotspot")).collect(Collectors.toList()));
    }
    // Convert radius to pixels
    double radiusPixels = radius / densityServer.getPixelCalibration().getAveragedPixelSize().doubleValue();
    try (@SuppressWarnings("unchecked") var scope = new PointerScope()) {
        for (var parent : parents) {
            ROI roi = parent.getROI();
            // We need a ROI to define the area of interest
            if (roi == null) {
                if (densityServer.nTimepoints() > 1 || densityServer.nZSlices() > 1) {
                    logger.warn("Hotspot detection without a parent object not supported for images with multiple z-slices/timepoints.");
                    logger.warn("I will apply detection to the first plane only. If you need hotspots elsewhere, create an annotation first and use it to define the ROI.");
                }
                roi = ROIs.createRectangleROI(0, 0, densityServer.getWidth(), densityServer.getHeight(), ImagePlane.getDefaultPlane());
            }
            // Erode the ROI & see if any hotspot could fit
            var roiEroded = RoiTools.buffer(roi, -radiusPixels);
            if (roiEroded.isEmpty() || roiEroded.getArea() == 0) {
                logger.warn("ROI is too small! Cannot detected hotspots with radius {} in {}", radius, parent);
                continue;
            }
            // Read the image
            var plane = roi.getImagePlane();
            RegionRequest request = RegionRequest.createInstance(densityServer.getPath(), downsample, 0, 0, densityServer.getWidth(), densityServer.getHeight(), plane.getZ(), plane.getT());
            var img = densityServer.readBufferedImage(request);
            // Create a mask
            var imgMask = BufferedImageTools.createROIMask(img.getWidth(), img.getHeight(), roiEroded, request);
            // Switch to OpenCV
            var mat = OpenCVTools.imageToMat(img);
            var matMask = OpenCVTools.imageToMat(imgMask);
            // Find hotspots
            var channels = OpenCVTools.splitChannels(mat);
            var density = channels.get(channel);
            if (minCount > 0) {
                var thresholdMask = opencv_core.greaterThan(channels.get(channels.size() - 1), minCount).asMat();
                opencv_core.bitwise_and(matMask, thresholdMask, matMask);
                thresholdMask.close();
            }
            // TODO: Limit to peaks
            if (peaksOnly) {
                var matMaxima = OpenCVTools.findRegionalMaxima(density);
                var matPeaks = OpenCVTools.shrinkLabels(matMaxima);
                matPeaks.put(opencv_core.greaterThan(matPeaks, 0));
                opencv_core.bitwise_and(matMask, matPeaks, matMask);
                matPeaks.close();
                matMaxima.close();
            }
            // Sort in descending order
            var maxima = new ArrayList<>(OpenCVTools.getMaskedPixels(density, matMask));
            Collections.sort(maxima, Comparator.comparingDouble((IndexedPixel p) -> p.getValue()).reversed());
            // Try to get as many maxima as we need
            // Impose minimum separation
            var points = maxima.stream().map(p -> new Point2(p.getX() * downsample, p.getY() * downsample)).collect(Collectors.toList());
            var hotspotCentroids = new ArrayList<Point2>();
            double distSqThreshold = radiusPixels * radiusPixels * 4;
            for (var p : points) {
                // Check not too close to an existing hotspot
                boolean skip = false;
                for (var p2 : hotspotCentroids) {
                    if (p.distanceSq(p2) < distSqThreshold) {
                        skip = true;
                        break;
                    }
                }
                if (!skip) {
                    hotspotCentroids.add(p);
                    if (hotspotCentroids.size() == nHotspots)
                        break;
                }
            }
            var hotspots = new ArrayList<PathObject>();
            int i = 0;
            for (var p : hotspotCentroids) {
                i++;
                var ellipse = ROIs.createEllipseROI(p.getX() - radiusPixels, p.getY() - radiusPixels, radiusPixels * 2, radiusPixels * 2, roi.getImagePlane());
                var hotspot = PathObjects.createAnnotationObject(ellipse, hotspotClass);
                hotspot.setName("Hotspot " + i);
                hotspots.add(hotspot);
            }
            if (hotspots.isEmpty())
                logger.warn("No hotspots found in {}", parent);
            else if (hotspots.size() < nHotspots) {
                logger.warn("Only {}/{} hotspots could be found in {}", hotspots.size(), nHotspots, parent);
            }
            parent.addPathObjects(hotspots);
        }
        hierarchy.fireHierarchyChangedEvent(DensityMaps.class);
        if (!toDelete.isEmpty())
            hierarchy.removeObjects(toDelete, true);
    }
}